Who decides what AI tells you? Campbell Brown, once Meta’s news chief, has thoughts

Forum AI, founded 17 months ago, raised $3 million led by Lerer Hippeau and is positioning itself as an AI evaluation company focused on high-stakes topics such as finance, hiring, geopolitics, and mental health. Campbell Brown said current models still show bias and factual errors, with her team aiming for roughly 90% consensus with domain experts to improve truthfulness and reliability. The article is largely a commentary on AI accuracy and compliance shortcomings rather than a direct market-moving event.

Analysis

META is a subtle loser here not because of a near-term revenue hit, but because the article sharpens a governance problem that sits at the center of Meta’s product and regulatory narrative: if AI-driven information flows remain noisy, biased, or context-poor, the company inherits more liability while capturing less trust premium. That matters because Meta’s consumer AI stack is likely to be judged on engagement first and epistemic quality second; if that order persists, the downside is not a single product miss but a compounding brand-and-policy discount that can cap multiple expansion over the next 6-18 months.

The bigger second-order winner may be the emerging AI evaluation/compliance layer, but the market will probably underwrite it as services rather than software until revenues become recurring. That creates a classic pick-and-shovel dynamic: model developers can talk about capability, but enterprises buying for lending, hiring, insurance, and security care about auditability and legal defensibility, which should pull spend toward benchmark, monitoring, and red-team tooling. The catch is procurement friction: most buyers will try to satisfy governance with cheap checkbox audits first, delaying a full re-rate of the category by at least 2-4 quarters.

The contrarian view is that the current skepticism toward AI may be structurally bullish for the companies that can prove control, not just power. If regulators or enterprise buyers shift from “can it answer?” to “can it be defended in court?”, the market will reward vendors that embed expert-guided evaluation into workflows and punish general-purpose model platforms with the highest error surface area. That transition is unlikely to show up in consensus estimates immediately, but it can drive faster sales cycles for governance tools while increasing legal and reputational risk for incumbent social and AI distribution platforms.

Key Decisions for Investors

• Underweight META on a 3-6 month horizon; use rallies to trim. Risk/reward skews negative if AI trust concerns bleed into ad-product and regulatory scrutiny, with limited near-term catalyst to re-rate higher.
• Long AI governance / compliance beneficiaries on weakness: build a basket in private-market proxies or public software names exposed to model monitoring, audit, and workflow control over the next 6-12 months. Upside comes from enterprises moving from checkbox audits to defensible evaluation spend.
• Pair trade: long enterprise AI infra/software with governance exposure, short consumer-facing AI distribution where trust is weakest. The spread should widen if regulatory headlines increase over the next 1-2 quarters.
• Sell out-of-the-money call spreads on META into any AI optimism spike. The thesis is that improved model quality is a slower-moving story than market enthusiasm, so implied upside can outpace realized governance progress.
• Set a catalyst watch for regulatory actions on AI audits and hiring/credit decision frameworks over the next 6-12 months; if enforcement tightens, expect a valuation uplift for evaluation vendors and a multiple haircut for platforms with opaque model behavior.